When AI helps match the egg with the best sperm!
The shape and general condition of the sperm are of crucial importance in the childbearing process. To improve observation techniques for these male gametes, a team of Monash University invented a learning strategy based on Artificial Intelligence. During testing, the algorithm had a diagnostic score better than that of human embryologists. The technique is currently being adapted for use in a clinical setting.
The process is based on several automatic models capable of individually classifying spermatozoa based on X-rays. Subsequently, an intelligent synthesis is carried out by combining the information collected.
Deirdre Zander-Fox, regional scientific director of Monash IVF and clinical expert, was one of the main contributors to the research. Researchers believe to achieve a beta prototype in 3 to 5 years.
A better understanding of sperm
Four different image classification models, called convolutional neural networks, were trained separately to group the sperm head. Each model learns, from a different angle, the relationship between the image of the sperm and its shape. The peculiarity here is the integration of parameters to improve test accuracy.
Additionally, another automatic approach called “metamodel” issues an executive decision based on the analyzes of the previous four models. Using this combination of processes, the scientists were able to a better understanding of sperm morphology during the assisted reproduction cycle.
” The real-time platform should be able to highlight sperm that have a better chance of fertilizing the egg. »
Reza Nosrati, researcher at Monash University
For better chances of success during inseminations
Despite the evolution of technologies for observing male gametes, the final decision rests with the attending physician. However, human intervention in this decision-making process would weaken diagnostic reliability. In addition, an inconsistency is observed between doctors’ diagnoses on the same patient.
The automatic model used therefore helps in the selection of the best sperm to use for insemination. Thus, the quality of the embryos and sperm sighting speed are improved. A unique feature of the method is the use of an ensemble approach, based on learning algorithms to obtain better predictive performance.
However, despite these promising results, the instrument should be improved to label and analyze all semen rather than the shape of the gametes alone. The integration of gamete motility and DNA integrity are also aspects to be taken into account.
” Embryologists do their best to classify sperm, but there is not enough technology at their disposal to help them automate this process. »
Reza Nosrati, researcher at Monash University
SOURCE: SCIENCE NEWS
